sequence number problem with zero length window size

[lkml@xxxxxxxxx]

I think I found a problem with IPv4/TCP sequence numbers when the
window length goes to zero.

This IS NOT the problem I'm reporting:

  We send data to a client that we believe has a problem.
  It acks packets but doesn't open the window size.
  Eventually, the window size goes to zero.

This IS the problem I'm reporting:

  When the window size goes to zero, the sender (Linux 2.4.19) sends a
  packet with the wrong sequence number.  In the attached trace, see
  packets starting at number 5470, time stamp 695.453571. Here the
  client acks for sequence number ....741 but the sender then sends
  a packet with sequence number ...740.

Looks like a bug.

Larry

trace2
Description: Binary data